ArticlePDF Available


Almost no one cites Sellars, while reinventing his wheels with gratifying regularity. (Dennett 1987, 349) In philosophy of mind, there is functionalism about mental states and functionalism about mental contents. The former — mental State functionalism — says that mental states are individuated by their functional relations with mental inputs, Outputs, and other mental states. The latter — usually called functional or conceptual or inferential role semantics — says that mental contents are constituted by their functional relations with mental inputs, Outputs, and other mental contents (and in some versions of the theory, with things in the environment). If we add to mental State functionalism the popular view that mental states have their content essentially, then mental state functionalism may be seen as a form of functional role semantics and a solution to the problem of mental content, namely, the problem of giving a naturalistic explanation of mental content. According to this solution, the functional relations that constitute contents are physically realized — in a metaphysically unmysterious way — by the functional relations between mental inputs, outputs, and the mental states bearing those contents.
Computationalism, and
Mental Contents1
Washington University
St. Louis, MO 63130-4899
Almost no one cites Sellars, while reinventing his wheels with
gratifying regularity. (Dennett 1987, 349)
In philosophy of mind, there is functionalism about mental states and
functionalism about mental contents. The former  mental state func-
tionalism  says that mental states are individuated by their functional
relations with mental inputs, outputs, and other mental states. The latter
 usually called functional or conceptual or inferential role semantics
 says that mental contents are constituted by their functional relations
with mental inputs, outputs, and other mental contents (and in some
versions of the theory, with things in the environment). If we add to
mental state functionalism the popular view that mental states have their
content essentially, then mental state functionalism may be seen as a
form of functional role semantics and a solution to the problem of mental
content, namely, the problem of giving a naturalistic explanation of
mental content. According to this solution, the functional relations that
constitute contents are physically realized in a metaphysically unmys-
terious way  by the functional relations between mental inputs, out-
puts, and the mental states bearing those contents. But for this solution
Volume 34, Number 3, September 2004, pp. 375-410
1 Id like to thank those who commented on previous versions of this paper, especially
Eric Brown, Robert Cummins, Frances Egan, Peter Machamer, Susan Schneider, and
two anonymous referees.
to be noncircular, the functional relations between mental inputs, out-
puts, and states must be specified in a way that does not appeal to the
contents of mental states.
Philosophers who endorse mental state functionalism typically also
endorse computationalism, or the Computational Theory of Mind
(CTM), according to which the functional relations between mental
inputs, outputs, and states are computational.2 Unfortunately, most of
the same philosophers also endorse what I call the semantic view of
computation, according to which computational relations are individu-
ated by the contents of the computational inputs, outputs, and internal
To distinguish it from mental content, I will call the content of com-
putational states computational content. The problem of how computa-
tional states have content, analogous to the problem of mental content,
may be called the problem of computational content. Since CTM ascribes
computations to the mind and the semantic view of computation holds
that computational states have their content essentially, the conjunction
of CTM and the semantic view of computation entails that CTM is a
representational theory, i.e. a theory that ascribes content to the mind.
Because of this, many computationalists have supposed that solving the
problem of computational content may be (perhaps part of) the solution
to the problem of mental content. Finally, the fact that CTM plus a theory
of computational content is seen as a theory of mental content is used as
a reason in favor of CTM: we should believe CTM because it offers (a
step towards) a naturalistic explanation of mental content.4
376 Gualtiero Piccinini
2 For an extended discussion of the relationship between these two doctrines, see
Piccinini forthcoming a.
3 Cf. Fodor: Ive introduced the notion of computation by reference to such semantic
notions as content and representation: a computation is some kind of content-re-
specting causal relation among symbols (Fodor 1998, 11). Cf. also Pylyshyn 1984,
4 This attitude is well expressed in the following passage, where the author seems to
see no relevant distinctions between semantic, computational, informational, and
intentional descriptions:
It is widely recognized that computation is in one way or another a symbolic or
representational or information-based or semantical  i.e., as philosophers
would say, intentional  phenomenon. Somehow or other, though in ways we
do not yet understand, the states of a computer can model or simulate or repre-
sent or stand for or carry information about or signify other states in the world....
The only compelling reason to suppose that we (or minds or intelligence) might
be computers stems from the fact that we, too, deal with representations, sym-
bols, meanings, and the like (Smith 1996, 9-11; emphasis added).
But the semantic view of computation generates a circularity between
functional role semantics, mental state functionalism, and computation-
alism. According to mental state functionalism plus the view that mental
states have their content essentially, mental contents are constituted by
the functional relations of the mental states that carry the contents.
According to computationalism, those functional relations are compu-
tational. And according to the semantic view of computation, computa-
tional relations are individuated by the contents of the states that enter
those relations. We are back to explaining contents, which is what we
were hoping to explain in the first place.
This paper tells the story of how contemporary philosophers of mind
entangled themselves in this circularity, so that computation and content
got almost inextricably intertwined, and how the same philosophers
tried to untangle themselves. The purpose is twofold: on one hand, to
reconstruct an episode of recent and influential philosophy; on the other
hand, to diagnose what went wrong. The upshot is that the semantic
view of computation should be replaced by a functional view of compu-
tation, and that the problem of mental content should be solved inde-
pendently of the question of whether mental states are computational. I
hope this will make room for a better understanding of both computation
and content.
I Content in Early Computationalism
The modern history of computationalism, and the connection between
computation and content, goes back to the origin of computability
theory. Alan Turing formulated his theory of computation in the context
of investigations on the foundations of mathematics.5 For instance, he
wanted to prove that the decision problem for first order logic had no
algorithmic solution. He formulated a theory of computation in terms of
what are now called Turing Machines (TMs). Turing argued that some
TM could carry out any computation that a human being could carry
out. In his argument, Turing described TMs anthropomorphically as
scanning their tape, seeing symbols, having memory or mental
states, etc., although he introduced all these terms in quotation marks,
presumably to underline their metaphorical use (Turing 1936-7, 117-18).
Moreover, in using TMs for his mathematical purposes, Turing assigned
interpretations to the inputs and outputs of TMs, usually as encoding
Functionalism, Computationalism, Mental Contents 377
5 For more details, see Piccinini 2003b and 2003a, chs. 1 and 3.
real numbers, but sometimes as encoding TM programs or formulae in
a logical calculus. So far, there was nothing methodologically problem-
atic with what Turing did. TMs were to be understood as mechanisms
for deriving strings of symbols, and the theorist was free to assign
interpretations to the strings (within the relevant methodological con-
In the 1940s, a few years after the publication of Turings theory,
stored-program digital computers were built. In an article that became
very influential, Turing argued that digital computers could be pro-
grammed to carry out conversations indistinguishable from conversa-
tions with humans (Turing 1950). In explaining digital computers to an
audience who was likely to know little about them, Turing used inten-
tional (and hence semantic) language again. He drew an analogy be-
tween digital computers and computing humans. The rules followed by
a human were analogous to the instructions stored by the computer, and
the human process of applying the rules was analogous to the com-
puters process of executing its instructions: It is the duty of the [com-
puters] control to see that these instructions are obeyed correctly and in
the right order (ibid., 437). Turings analogy helped explain succinctly
what digital computers did, and for that purpose, there was nothing
objectionable to it. But it had the potential to suggest that computers
somehow understood and obeyed instructions similarly to how people
understood and obeyed instructions. This analogy, by no means limited
to Turings writings, was a likely source of the semantic view of compu-
tation.6 When combined with CTM, the semantic view of computation
would be used to generate theories that seemed to explain mental
The modern form of CTM was formulated by Warren McCulloch in
the 1930s and published by him in the 1940s.7 McCulloch held that the
378 Gualtiero Piccinini
6 For example, later we will examine writings by Jerry Fodor, who was an important
early proponent of the semantic view of computation. In his early work on this
matter, Fodor maintained that computational descriptions are semantic, and the
primary reason he gave was that computers understood and executed instruc-
tions (Fodor 1968a; 1968b, 638). Only later did he add that computers operated on
representations (Fodor 1975).
7 CTM is often attributed to Turing (e.g., by Fodor 1998). Although Turing occasion-
ally wrote that the brain was a computer (see the essays collected in Ince 1992), his
statements to that effect were made after McCullochs theory, which Turing knew
about, had been published (McCulloch and Pitts 1943). I dont know of any place
where Turing stated that thinking is computation. In fact, Turing denied that
intelligence and thinking were theoretically useful concepts (Turing 1950).
McCulloch, however, explicitly held the view that thinking was computation (see
brain was a computing mechanism and that thinking was computation.
He also argued that CTM explained the possibility of human knowledge
and solved the mind-body problem. During the 1940s, CTM was
adopted and elaborated by Norbert Wiener, John von Neumann, and
other members of the newly forming cybernetics community, one of
whose goals was to explain the mind by building computational models
of the brain. During the 1950s, students and younger colleagues of
McCulloch, Wiener, and von Neumann turned CTM into the foundation
of the new discipline of artificial intelligence, whose goal was to explain
the mind by programming computers to be intelligent.8
Early computationalists described computers and neural mechanisms
using semantic language. For instance, they said that computers (either
neural or artificial) manipulated numbers, suggesting that something
in the computer meant or represented numbers:
Computing machines are essentially machines for recording numbers, operating
with numbers, and giving the result in numerical form (Wiener 1948, 137).
Existing computing machines fall into two broad classes: analog and digital.
This subdivision arises according to the way in which the numbers, on which the
machine operates, are represented in it (von Neumann 1958, 3; emphasis added).
Th us th e n erv ou s s ys tem ap pe ar s to be us in g a r ad ic al ly d if fe re nt s ys te m o f no ta ti on
from the ones we are familiar with in ordinary arithmetics [sic] and mathematics:
instead of the precise systems of markers where the position  and the presence or
absence  of every marker counts decisively in determining the meaning of the
message, we have here a system of notations in which the meaning is conveyed by
the statistical properties of the message (von Neumann 1958, 79; emphasis added).
Early computationalists talked about computers having content, but
they did not discuss how this was possible. They were concerned with
building machines that exhibited intelligent behavior, not with philo-
sophical issues about content. Accordingly, they did not address the
problem of computational content explicitly. They did not even say
explicitly whether they thought computers and minds had content in the
same sense, although their writings give the impression that they
thought so. McCulloch, for example, argued that CTM explained human
Functionalism, Computationalism, Mental Contents 379
e.g. the essays collected in McCulloch 1965). For more on Turings views on
intelligence, see Piccinini 2003b.
8 A detailed analysis of McCulloch and Pittss theory is in Piccinini, forthcoming b.
For more details on the early history of computationalism, see Piccinini 2003a, chs.
2 through 6.
9 For a more comprehensive discussion of von Neumanns views on the computer
and the brain, see Piccinini 2002a.
knowledge, which suggests that the content of the computational states
postulated by CTM explained the content of knowledge states. But he
did not discuss how a computational state acquired its content or how
this related to the content of mental states.
Pa rt of t he rea so n fo r t his la ck o f i nter est i n the p roble m of c onten t may
be due to a certain operationalist spirit that early computationalists
The science of today is operational: that is, it considers every statement as essentially
concerned with possible experiments or observable processes. According to this, the
study of logic must reduce to the study of the logical machine, whether nervous or
mechanical, with all its non-removable limitations and imperfections. (Wiener 1948,
The problem of giving a precise definition to the concept of thinking and of
deciding whether or not a given machine is capable of thinking has aroused a great
deal of heated discussion. One interesting definition has been proposed by A.M.
Turing: a machine is termed capable of thinking if it can, under certain prescribed
conditions, imitate a human being by answering questions sufficiently well to
deceive a human questioner for a reasonable period of time. A definition of this type
has the advantages of being operational or, in the psychologists term, behavioristic.
No metaphysical notions of consciousness, ego and the like are involved. (Shannon
and McCarthy 1956, v)
This operationalism may have led early computationalists to discount
questions about computational or mental content on the grounds that
either content could not be operationalized, or it had to be operational-
ized in non-semantic terms. Be that as it may, early computationalists
formulated their CTM using semantic language, but they had no theory
of content, and they gave no indication that they thought they needed a
theory of content. This is probably the origin of the view that CTM
ascribes content to the mind, and that it has something to contribute
towards solving the problem of mental content.11
380 Gualtiero Piccinini
10 Notice that Turing 1950 did not offer a definition of intelligence, let alone an
operational one (cf. Moor 2001, Piccinini 2000). In reading Turing as offering an
operational definition of intelligence, Shannon and McCarthy showed how strong
their own operationalist leanings were.
11 I believe that the cybernetic form of CTM, formulated using semantic language, later
spread (in the same semantic form) into AI, psychology, and philosophy. I will not
document this thesis here (but see Piccinini, forthcoming a, for its influence on
Hilary Putnam).
II Functional Role Semantics
Before discussing computationalism in philosophy, I should mention the
theme of content in the philosophy of language. In the 1950s through the
1970s, many philosophers were led to think about mental content by
thinking about linguistic content and attempting to formulate a semantic
theory of language. By semantic theory, I mean both an assignment of
content to language (a semantics) and an account of how language
acquires its content. For present purposes, the second question is the
important one.12
A natural way to account for linguistic content is to say that it comes
from the content of the language users minds. In other words, linguistic
content can be explained by postulating that certain mental states of the
language users have appropriate contents that are transferred to the
speakers utterances. But postulating mental content as an explanation
for linguistic content calls, in turn, for a theory of mental content.
A possible solution to the problem of mental content is a theory that
goes back Wilfrid Sellars (1954, 1956, 1961, 1967, 1974). In Sellarss
theory, mental states  more specifically, thoughts were construed
by analogy with linguistic sentences, and mental contents were consti-
tuted by the relations of individual thoughts to stimuli (linguistic and
behavioral), responses (linguistic and behavioral), and other thoughts.
Each thought was individuated by its content, and its content was
constituted by on the one hand the inputs and other thoughts that elicited
that thought, and on the other hand the outputs and other thoughts that
were elicited by it.13 Because of its reliance on inner linguistic episodes,
Sellarss theory was said to postulate a language of thought (Harman
1970, 404). Different authors describe the role played by mental states in
this sort of theory as either functional, or inferential, or conceptual, and
the resulting theory of content is correspondingly called functional, or
inferential, or conceptual role semantics.
Functionalism, Computationalism, Mental Contents 381
12 On the history of philosophical discussions on content in American philosophy from
the 1950s on, including many themes that I have no room to mention here, see
Harman 1968, 1988; and especially Dennett 1987, ch. 10.
13 Ludwig Wittgenstein (1953) offered a similar theory of linguistic content, without
extending it to the content of mental states. Sellarss view appears to have originated
independently of Wittgensteins. In his Autobiographical Reflections, Sellars
traced his functionalism about content back to reflections that he made in the 1930s
and the subsequent influence of Immanuel Kant on him (Sellars 1975, 285-6; see also
Sellars 1974, 463).
With functional role semantics in place, we can go back to computa-
III Computationalism and the Philosophy of Mind
Computation became an important notion in contemporary philosophy
of mind through work by Hilary Putnam and his student Jerry Fodor in
the 1960s (Putnam 1960, 1963, 1964, 1967a, 1967b; Fodor 1965, 1968a,
Putnam was familiar with some of the cybernetics literature, including
McCullochs CTM,15 and like the cyberneticians, he did not seem con-
cerned with formulating a theory of mental content. In the first paper
where he drew an analogy between minds and TMs (Putnam 1960),
Putnam introduced computational descriptions to dissolve the mind-
body problem. He argued that a problem analogous to the mind-body
problem arose for TMs, and that this showed the mind-body problem to
be a purely verbal or linguistic problem. In the same paper, Putnam
said that internal states of TMs were individuated by their functional
relations to inputs, outputs, and other internal states. Two years later
(Putnam 1967a, delivered to the Wayne State University Symposium in
the Philosophy of Mind in 1962), Putnam argued that mental states were
individuated in the same (functional) way that TM states were, but left
open the question of whether minds were TMs or something more
complicated. In this occasion, he offered his analogy between minds and
TMs as a solution to (as opposed to a dissolution of) the mind-body
problem. Putnams mind-body doctrine was going to be called (mental
state) functionalism. In its canonical formulation, called computational
functionalism, it stated that minds were, in fact, a kind of TM (Putnam
382 Gualtiero Piccinini
14 The relationship between Putnam and Fodors computationalism and their mental
state functionalism, which pertains only indirectly to the present topic, is explored
in more detail in Piccinini, forthcoming a.
15 McCulloch and Pittss theory is cited both in Oppenheim and Putnam 1958 and in
Putnam 1964. So Putnam knew McCulloch and Pittss theory before moving to MIT
(where McCulloch was) in 1961. During the early 1960s, both Putnam and Fodor
were at MIT, which was perhaps the main center of cybernetics research as well as
the home institution of Noam Chomsky, who was proposing to explain the human
ability to manipulate language by postulating innate knowledge of a recursive (i.e.
computational) grammar (Chomsky 1957, 1965). At that time, both Putnam and
Fodor were close to Chomsky and his views (Putnam 1997).
Fodor was influenced by Putnam and by psychologists J.A. Deutsch
(1960) and Stuart Sutherland (e.g., Sutherland 1960; Fodor 1965, 161, and
personal correspondence). One of his goals was an account of psycho-
logical explanation. In his first paper on this subject, Fodor wrote that
psychological theories were functional analyses, i.e. theories that de-
scribed a system in terms of internal states and state transitions that were
specified by their functional role (Fodor 1965). Later, Fodor added that
psychological functional analyses were lists of instructions, or programs,
which explained behaviors by stating which internally stored instruc-
tions were executed by people engaged in those behaviors (Fodor
Although in these writings Putnam and Fodor did not seem concerned
directly with the problem of mental content, Putnam and Fodors 1960s
papers on the metaphysics of the mind overlapped considerably with
Sellarss theory of mental content. The main common theme  ex-
pressed by different authors using different terminologies  was that
mental states were to be individuated by their functional relations within
a network of inputs, outputs, and other mental states. Another theme
was the use of functionalism to dispense with arguments from the
privacy of the mental to some special ontological status of the mental
(Sellars 1954; Putnam 1960). Finally, there was the analogy between
minds and computers. In this respect, Sellars wrote:
[The] learning of a language or conceptual frame involves the following logically
(but not chronologically) distinguishable phases:
(a) the acquisition of S[timulus]-R[esponse] connections pertaining to the arranging
of sounds and visual marks into patterns and sequences of patterns. (The acquisition
of these habits can be compared to the setting up of that part of the wiring of a
calculating machine which takes over once the problem and the relevant infor-
mation have been punched in.)
(b) The acquisition of thing-word connections. (This can be compared to the setting
up of that part of the wiring of the machine which enables the punching in of
information.) (Sellars 1954, 333)
Despite the overlap between Sellarss functionalism in the 1950s and
Putnam and Fodors functionalism in the 1960s, there is little evidence
that Sellars influenced Putnam and Fodor.
Sellars was already well known and one of his early papers, Empiri-
cism and the Philosophy of Mind (Sellars 1956), was widely read and
Functionalism, Computationalism, Mental Contents 383
16 According to Harman (personal correspondence), this view of psychological theo-
ries was influenced by the view of some psychologists, who proposed a similar
vision of psychological theories to replace the behaviorist stimulus-response view
of psychological theories (Miller, Galanter, and Pribram 1960).
discussed when it came out (Harman, personal correspondence). By the
early 1960s, Putnam knew Sellarss paper (personal correspondence),
and later named it one of the most important papers on [its] topic in
recent decades (Putnam 1974, 445).17 But Putnam does not recall know-
ing Sellarss 1954 essay, in which Sellars explicitly defended his function-
alism about content (personal correspondence). Although Sellars 1956
did present his view that thoughts were analogous to inner linguistic
episodes as well as his functionalist theory of content (e.g., Sellars 1956,
180), functionalism about content was not the primary focus of that
essay. So although Putnam eventually learned about Sellarss theory,
initially he may not have seen it as a theory of content. As to Fodor,
according to him at that time he was not acquainted with Sellarss work
(personal correspondence). At any rate, neither Putnam nor Fodor cited
Sellars in their 1960s papers.
384 Gualtiero Piccinini
17 Dennett wrote that [it] is clear that Putnam[s functionalism has] ... been quite
directly influenced by Sellars (Dennett 1987, 341). Dennett told me he got his sense
of Putnams debt to Sellars during a discussion of Sellarss views with Putnam, a
discussion that took place in March 1973 (personal correspondence). Putnam,
ho we ver , to ld me h e  arr ive d a t fu nc tio nal is m qu it e na tur al ly, be ing at the tim e bo th
a philosopher and a recursion theorist, and Sellars 1956 didnt inspire my func-
tionalism (personal correspondence). Speaking of his work in the late 1950s, he also
wrote as follows (recall that, as we saw in section I, Turing had used mentalistic
language to describe his TMs):
I was in the habit of explaining the idea of a Turing machine [n. omitted] in my
mathematical logic courses in those days. It struck me that in Turings work, as
in the theory of computation today, the states of the imagined computer (the
Turing machine) were described in a very different way than is customary in
physical science. The state of a Turing machine  one may call such states
computational st at es  is id en ti fie d b y i ts ro le i n c er tai n c om pu tat io na l p roc es se s,
independently of how it is physically realized. A human computer working with
paper and pencil, a mechanical calculating engine of the kind that was built in
the nineteenth century, and a modern electronic computer can be in the same
computational state, without being in the same physical state. I began to apply
images suggested by the theory of computation to the philosophy of mind, and
in a lecture delivered in 1960 [n. omitted; the lecture was published as Putnam
1960] I suggested a hypothesis that was to become influential under the name
functionalism: that the mental states of a human being are computational states
of the brain (Putnam 1997, 180-1).
Curiously, here Putnam followed a common pattern in the literature on functional-
ism, which attributes computational functionalism to Putnam 1960 even though
Putnam didnt formulate it until Putnam 1967b. Putnam 1960 did formulate an
analogy between minds and TMs, but also denied that minds could be characterized
in functional terms in the way TMs were. Perhaps reading Empiricism and the
Philosophy of Mind contributed to Putnams shift from an anti-functionalist posi-
tion in his 1960 paper to his later functionalist position, but there is no direct
evidence of this.
The development of Putnam and Fodors functionalism appears to
have been largely independent of Sellarss functionalism. If this is true,
it helps explain why Putnam and Fodor did not discuss the problem of
mental content, did not distinguish between the classical mind-body
problem and the problem of mental content, and did not seem to think
that after giving their (computational) functionalist solution to the mind-
body problem, the problem of mental content remained to be solved.18
IV The Semantic View of computation
in the Philosophy of Mind
Another reason for ignoring the problem of mental content may be that
Putnam and Fodor formulated their functionalist theory of mental states
using computational descriptions, and  like the cyberneticians 
Putnam and Fodor individuated computational states using semantic
In the papers cited above, Putnam was ambivalent on computational
states and content. In his 1960 paper, he drew an analogy between mental
states and TM states, and between introspective reports and TM self-de-
scriptions, but he added that TMs could not be properly said to use a
language. Later (Putnam 1967b), he stated that mental states were TM
states. Together with the generally shared premise that some mental
states have content, which Putnam never rejected, this entailed that at
least some TM states had content.
Fodor argued that computational explanations in psychology were
intellectualist in Ryles sense (Ryle 1949). An intellectualist explanation
accounted for an intentionally characterized overt behavior by postulat-
ing an intentionally characterized internal process. Ryle criticized intel-
lectualist explanations; roughly speaking, he argued that they required
the postulation of an internal homunculus to explain the intentionally
characterized internal process, thereby generating an infinite regress of
homunculi inside homunculi (Ryle 1949). Fodor (1968b) rejected Ryles
criticism of intellectualist explanations on the grounds that computers
activities were characterized in intellectualist terms but involved no
Functionalism, Computationalism, Mental Contents 385
18 A fortiori, they did not discuss whether mental states have their content essentially,
a view that is popular nowadays. If one believes that mental states have their content
essentially, then one will automatically see Putnams and Fodors early functional-
ism as providing a theory of content. Otherwise, one will interpret their work as
offering a theory of the identity conditions of mental states that is neutral about their
content (cf. Jackson and Pettit 1988, 388).
infinite regress. Fodor built an explicit view about computation ascrip-
tion around the idea that computational descriptions ascribed semantic
properties. According to Fodor, for something to be a computing mecha-
nism, there must be a mapping between its physical states and certain
intellectualistic (hence semantic) descriptions of what it does:
[A] programming language can be thought of as establishing a mapping of the
physical states of a machine onto sentences of English such that the English sentence
assigned to a given state expresses the instruction the machine is said to be executing
when it is in that state. (Fodor 1968b, 638, emphasis added)
Every computational device is a complex system which changes physical state in
some way determined by physical laws. It is feasible to think of such a system as a
compute r just i nsofar as it is p ossible to d evise some m apping which pairs physical
states of the device with formulae in a computing language in such a fashion as to
preserve desired semantic relations among the formulae. (Fodor 1975, 73; emphasis
These passages show that for Fodor, computational descriptions as-
cribed semantic properties to a mechanism and individuated its states
by reference to those semantic properties. This was the semantic view of
The semantic view is not a theory of computational content. A theory
of content explains how something acquires its content. The semantic
view of computation is simply the view that computational states do
have content (essentially)  that describing something as a computing
mechanism is a way of ascribing content to it. The semantic view of
computation does not specify by virtue of which properties or conditions
computational states acquire their putative content.19
The blending of computational functionalism, CTM, and the semantic
view of computation culminated in Fodors Language of Thought (LOT)
hypothesis and his famous slogan no computation without repre-
sentation (Fodor 1975). According to Fodor 1975, learning what a predi-
cate in a public language meant required representing the semantic
properties of that predicate (e.g., the predicates extension) in some
previously understood language LOT. Now, if understanding LOT
386 Gualtiero Piccinini
19 In order to have a theory of content, at the very least one needs to add to the semantic
view of computation an interpretational semantic theory, according to which all
there is to having content is being appropriately described as having content. As
an anonymous referee has pointed out to me, the above passages from Fodor 1968b
and 1975 could be read as implicitly suggesting such an interpretational semantics.
Given Fodors realism about content, he would probably reject such a reading, but
this is beside the point. Interpretational semantics is discussed in section VII below,
and Fodors alternative to it is discussed in section VIII.
required representing the semantic properties of its predicates in some
previously understood language, this would lead to infinite regress.
Fodor blocked this infinite regress by appealing to stored-program
computers and the way they responded to their inputs and instructions.20
Modern computers received data and instructions as inputs written in
some programming language and then transformed those inputs into
machine language code that they could execute. But executing code did
not require a new transformation of internal code into another language,
or to possess a representation of how to carry out the execution. Com-
puters were hardwired to carry out certain elementary operations in
response to certain lines of internal code, so the regress stopped at those
hardwired processes (Fodor 1975, 65ff.).
Fodor likened human public languages to high level programming
languages, and the human LOT to a computers machine language. He
argued that LOT was our best explanation for human cognition, and
specifically for the human ability to manipulate language and make
inferences in a way that respected the semantic properties of thoughts.21
As we saw above, Fodor described computers and their languages using
semantic idioms, perhaps in part because his appeal to stored-program
computers was intended to render LOT mechanistically intelligible. So
although LOT explained the content of language and mental states in
terms of the content of LOT expressions, it did not include a theory of
the content of the LOT expressions themselves. To be hardwired to
execute certain lines of code was a very interesting property that stored-
Functionalism, Computationalism, Mental Contents 387
20 Fodor also argued that LOT was innate. This component of Fodors version of LOT,
which was rejected by many who accepted LOT, is irrelevant to the present
21 This was one more similarity with Sellarss ideas, which included the postulation
of inner linguistic episodes as an explanation for thought (esp. Sellars 1956). Sellarss
ideas were turned into a systematic theory by Harman (1973, discussed below).
Fodor told me he learned about Sellarss and Harmans theory of thought only after
writing his 1975 book (personal correspondence). He added that around the same
time Zeno Vendler also wrote a book that proposed a similar theory of thought
(Vendler 1972). Although Vendler preferred not to talk of an inner language (ibid.,
42, 51), his theory postulated an innate neural code that could be scientifically
deciphered (ibid., 142). According to Vendler, he developed his theory by trying to
improve on Austins theory of illocutionary acts under the influence of Chomskian
linguistics (ibid., viii, 4). Vendler did not refer to Sellarss theory of thought.
Although Fodor told me he recalls no mutual influences on their respective versions
of LOT between himself, Harman, and Vendler (personal correspondence), Fodor
did cite Vendler 1972s arguments as relevant to Fodors own argument for LOT
(Fodor 1975, 58, n. 4), and Harman remembers that Vendler attended a presentation
of Harmans version of LOT in 1968 (personal correspondence).
program computers had, but it could not be identified with having
content, much less having content corresponding to the content of hu-
man language and thought, without argument. Fodor 1975 did not
address how the process of translation between public language and
LOT respected the semantics of the public language, namely how LOT
acquired its semantics and managed to match it with the semantics of
the public language. Unlike Sellarss LOT, Fodors LOT offered no
solution to the problem of content, nor did he purport to offer such a
During the 1970s, CTM became very influential in philosophy of mind.
Many philosophers accepted some version of CTM, even though some
of them rejected one or another tenet of Fodors LOT. At the same time,
we will see that the main authors who discussed CTM sympathetically
subscribed to the semantic view of computation. They thought the mind
was computational, and computational states had content, so they
thought the problem of mental content might be reducible to the problem
of computational content. And since computing mechanisms were
mechanisms built by humans, the problem of computational content
may have seemed less philosophically pressing than the problem of
mental content. Nevertheless, solving the problem of mental content
requires combining CTM with a theory of content.
V Computationalism and Theories of Content
In the rest of this paper, I will argue that computation ascription alone
is insufficient for the ascription of mental content. If there were consen-
sus about what mental content is, I could run a general argument of the
following form:
Premise 1. Having mental content is the same as satisfying condi-
tion C.
Premise 2. Being a computational state does not entail satisfying
condition C.
388 Gualtiero Piccinini
22 Nor did he make explicit that his theory presupposed such a solution. I insist on
th is b ec aus e L OT h as b ee n so met im es m ist ak en f or a th eor y o f co nte nt . Fo r ex am ple ,
Putnam criticized Fodors LOT as if it were a theory of content (Putnam 1988, 21,
40-1). Perhaps this says something about how Putnam was thinking about compu-
tation and content. I took this reference to Putnam from Loewer and Rey, who also
point out t hat LOT i s not a theo ry of content (Loe wer an d Rey 1991, xix) . The rea son
why Fodors LOT is misread as a theory of content may be partially due to how
Fodor blended it with the semantic view of computation.
Conclusion: Being a computational state does not entail having
mental content.
Since there is no consensus on what mental content is, I will go through
the main theoretical approaches to content and argue that in each case,
ascribing any kind of content to computational states presupposes a
non-semantic individuation of the computational states.
A consequence is the rejection of the thesis that CTM contributes to
solving the problem of mental content, which eliminates this as a reason
for believing CTM. My conclusion has no consequences on whether
minds or computing mechanisms have content, whether mental and
computational content are the same, and the project of reducing mental
content to computational content. All I argue is that those questions must
be answered by a theory of content, not by a theory of computation or a
VI CTM meets Functional Role Semantics
Among computationalist philosophers, the first who took the problem
of mental content seriously was probably Gilbert Harman. Harman was
familiar both with Putnam and Fodors computational functionalist
writings and with Sellarss theory of content (Harman 1968, 1970).23 His
idea was to explicitly combine computational functionalism about men-
tal states with a Functional Role Semantics (FRS) about their content, so
as to have a naturalistic theory of semantically individuated mental
states. According to FRS, the content of mental states was constituted by
their functional relations. But FRS, as Sellars left it, did not specify a
mechanism that could physically realize those functional relations. From
a naturalistic perspective, a FRS for mental states called for a theory of
the mechanism realizing the functional relations. Computational func-
tionalism had a mechanism that seemed to have the properties needed
to realize the relevant functional relations  after all, at least under the
semantic view of computation, computing mechanisms draw inferences
 while at the same time lacking a theory of the mechanisms content.
By combining the two, Harman (1973) could use the one theory to
solve the problem left open by the other, and vice versa. On one hand, he
appealed to the roles of computational states within a computing mecha-
nism as appropriate realizers of the functional relations that ascribed
Functionalism, Computationalism, Mental Contents 389
23 In personal correspondence, Harman has added to these the influences of Miller,
Galanter, and Pribram 1960; and Geach 1956.
content to mental states (ibid., 43-8). On the other hand, he appealed to
those functional relations to ascribe content to the states of the mecha-
nism (ibid., 60). The result was a CTM in which the computational states
and processes that constituted the mind were also the physical realizers
of the functional relations that gave the mind its content. In the same
theory, Harman also maintained Sellarss construal of thoughts as analo-
gous to linguistic sentences, i.e., as internal representations.
After Harmans revival of it, FRS found many advocates. Some ac-
cepted Harmans combination of FRS and a representational CTM (e.g.,
Field 1978). In a similar vein, Ned Block argued that FRS was the best
available theory of mental content to combine with LOT (Block 1986).
Others took from Harman only FRS and a functional individuation of
mental states, while discarding LOT (e.g., Loar 1981). And Paul
Churchland, a student of Sellars, developed a version of FRS that was
not conjoined with any form of CTM (Churchland 1979), although later
Churchland embraced a connectionist version of CTM (Churchland
For present purposes, it is important to understand the division of
labor in the combined CTM-FRS theory. The content of a mental state
comes from its functional relations, so whether the component of the
theory that accounts for mental content is successful or not depends on
whether functional relations are adequate to provide the semantics of
mental states. The functional relations of a mental state are (partially)
individuated by the computational relations that the state bears to other
states, inputs, and outputs.24 So, the computational states and relations
must be individuated in a way that does not presuppose their content,
otherwise the theory of content becomes circular. In other words, a
combination of CTM and FRS individuates content by functional rela-
tions, and functional relations (at least in part) by computational rela-
tions. If the computational relations are individuated by appeal to
content, as the semantic view of computation would have it, then CTM-
FRS is running in a small circle. If CTM-FRS wants to avoid circularity,
it needs a non-semantic view of computation, i.e. a way to individuate
390 Gualtiero Piccinini
24 CTM-FRS theorists do not individuate content itself with computational roles,
however. Some, like Harman, postulate that the relations between inputs and
outputs on one hand, and the environment on the other, also contribute to the
relevant functional relations and hence to content (broad FRS); others, like Block,
postulate that functional relations are only one of the factors determining content,
the other being reference (narrow FRS). These subtleties make no difference for our
purposes; see Block 1986 for discussion.
computational states and their relations without appeal to their con-
As natural as it may seem to combine CTM and FRS, it is not manda-
tory. The two are logically independent. Even if one believes that content
is constituted by functional relations, the physical realization of the
functional relations need not be computational.26 And even if one be-
lieves that the mind is computational and that computational states have
content, one need not believe that content is constituted by the functional
relations of the computational states. For a first important alternative
theory of content for minds and computing mechanisms, we will look at
the tradition that follows Daniel Dennett.
VII CTM meets Interpretational Semantics
Like Harman, Dennett was familiar with both computational function-
alism and some of Sellarss work, and he was interested in the problem
of mental content. Dennett also continued the tradition of his mentor
Gilbert Ryle, whose analytical behaviorism rejected the postulation of
semantically characterized mental states (Ryle 1949). Following Sellars
and Putnam, Dennett did develop a version of functionalism about
content, according to which mental content was constituted by the
mutual relations between contentful states, inputs, and outputs (Dennett
1969, ch. 4).27 But Dennetts functional relations between contents, unlike
Functionalism, Computationalism, Mental Contents 391
25 I dont know of any place where CTM-FRS theorists have offered such an account.
Block (1986) says he doesnt know how to individuate functional relations. To a
direct question, Harman answered thus: I dont think I have ever tried to provide
a noncircular theory of content in that sense (personal correspondence).
26 To say more about what a non-computational realization of functional relations
would be would take us too far afield. Given my account of computing mechanisms
in Piccinini 2003a, ch. 10, FRS could be combined with a non-computational func-
tional analysis of the mind-brain, i.e. a functional analysis that explains cognitive
processes without ascribing computations to the mind-brain. For versions of FRS
that are not committed to computational roles as physical realizers of functional
relations, see Peacocke 1992, and Brandom 1994.
27 It seems that Dennetts functionalism about content was more indebted to Putnams
computational functionalism than to Sellarss FRS. Dennett put it as follows:
I had read some Sellars when I finished Content and Consciousness [Dennett
1969], but I hadnt thought I understood it very well. Several of his students had
been in Oxford with me, and had enthused over his work, but in spite of their
urging, I didnt become a Sellarsian. Id read all of Putnams consciousness pa-
pers (to date), and was definitely influenced strongly by Putnam. One of Sellars
Harmans, were not realized by computational roles (Dennett 1969,
Following Ryle, Dennett argued that explaining content by postulat-
ing contentful mental states was tantamount to postulating a homuncu-
lus who understood the content of the mental states. This, however,
either begged the question of how the homunculus understands content,
or led to the infinite regress of postulating homunculi inside homunculi.
Dennett argued that mental states and their contents came from the
external observer of a system; they were ascribed to people, not discov-
ered in them, by describing and predicting the behavior of a system from
what Dennett called the intentional stance. According to Dennett, people
ascribed contentful states, e.g. beliefs and desires, to each other in order
to predict each others behavior, but inside peoples brains there was
nothing that realized those contentful states in an exact way, as Har-
mans computational states were supposed to do (Dennett 1987, 53, 71).
If a systems behavior was sufficiently complex and adaptive and an
external observer did not know its internal mechanisms well enough to
derive its behavior mechanistically, then the observer interpreted the
system as possessing beliefs and desires about its environment  hence
as having contentful states  so as to explain the systems adaptive
behavior as the satisfaction of appropriate desires given mostly true
beliefs. In summary, an observer ascribed content to the system. So
mental content came from the interpretation of external observers.
A few years after Dennett formulated his theory of content, Fodors
LOT hypothesis convinced Dennett that content could be ascribed to
internal states of a system, and specifically to states of stored-program
computers, without begging the question of how content was under-
stood (Dennett 1978). Dennett explained why the question was not
begged by applying his interpretational semantics to the internal states,
in a version of his theory that was later called homuncular functionalism
(ibid.). According to homuncular functionalism, content (and the intel-
ligence needed to understand it) could be ascribed to the internal states
392 Gualtiero Piccinini
students, Peter Woodruff, was a colleague of mine at UC Irvine, and it was he
who showed me how my work was consonant with, and no doubt somewhat
inspired by, Sellars. But that was after he read C&C. I thereupon sent Sellars one
of the first copies of C&C, and he wrote back enthusiastically....
I would think that my sketchy functionalist theory of meaning was more influ-
enced by Putnams Minds and Machines paper [Putnam 1960] than anything
Id read in Sellars while at Oxford, but I cant be sure. I have sometimes discover-
ed telltale underlinings in my copy of a book years later and recognized that I
ha d bee n in flu enc ed b y an aut hor and utt erly forgotten it (Dennett, personal cor-
of a system without begging the question of how it was understood as
long as the ascribed content was ultimately discharged by a completely
mechanistic explanation of the behavior of the system. Discharging
content ascription mechanistically consisted of decomposing the system
into parts and explaining the content and intelligent behavior of the
system in terms of the content and intelligent behavior of the parts. In
doing this, the parts and their behavior could still be ascribed content,
but understanding their content must require less intelligence than the
intelligence required by the content ascribed to the whole. The same
process could be repeated for the content ascribed to the parts, explain-
ing their content and intelligence in terms of the content and intelligence
of their parts, but the parts parts intentional descriptions must ascribe
less and less content and intelligence to them. The process ended with
components whose behavior was so simple and obviously mechanical
that it warranted no content ascription. Dennett offered his homuncular
functionalism as a way to cash out the intentional descriptions com-
monly used for computers and their programs, especially in the context
of artificial intelligence research. Thus, homuncular functionalism expli-
cated the semantic view of computation in a metaphysically non-myste-
rious way. It dissolved, rather than solve, the problem of computational
Dennetts theory went through elaborations and revisions (Dennett
1978, 1987), but its core remained: content, whether mental or computa-
tional, came from the interpretations of external observers. Dennetts
theory had the great advantage of being equally applicable to organisms
and artifacts like computers, giving a common treatment of computa-
tional and mental content. This was because both organisms and artifacts
were equally subject to intentional interpretation by external observers.
However, interpretational semantics also denied the reality of mental
content: for Dennett, mental content was an instrument of prediction; as
much as the predicted behavioral patterns were objective (Dennett 1987,
15, 25), the states posited by the predictive tool did not correspond to
any of the internal states posited by a correct mechanistic explanation of
the system whose behavior was being predicted. Put another way,
Dennett did not believe in original or intrinsic intentionality (e.g., Den-
nett 1987, 288). A similar point applied to computational content: when
Functionalism, Computationalism, Mental Contents 393
28 In formulating homuncular functionalism, Dennett also argued that the Church-
Turing thesis entailed that a mechanistic theory had to be computational, hence that
AI is the study of all possible modes of intelligence (Dennett 1978, 83). I discuss
this further component of Dennetts homuncular functionalism, which makes no
difference to the present discussion, in Piccinini 2003a, ch. 7.
computational content was discharged by the decomposition of the
system into the activity and internal states of purely mechanical (non-
contentful) components, computational content was explained away.
Furthermore, interpretation was somewhat indeterminate: in principle
the same behavior by the same system could be interpreted in different
and equally adequate ways (e.g., Dennett 1987, 40, ch. 8). Dennetts
theory explained content at the cost of deeming it unreal.
A corollary of Dennetts conjunction of the semantic view of compu-
tation and interpretational semantics is interpretationism about compu-
tation: whether something is a computing mechanism is a matter of
interpretation not fact. From the perspective of Dennetts intentional
stance, there is no principled difference between a desktop computer and
a refrigerator, except that applying the intentional stance to the computer
is more useful than applying it to a refrigerator. This leads to the
paradoxical effect that Dennetts intentional stance, which seems to
account so well for computational content, does not account for our
practice of applying the term computer only to some machines, which
seem to belong to a special class distinct from other machines. Given
Dennetts theory, in order to explain the difference between computing
mechanisms and other machines, one must abandon the intentional
stance and take some other stance (perhaps the design stance) towards
the mechanisms. A similar problem arises in comparing different inter-
pretations of the same computation. Given that for Dennett interpreta-
tion is partially indeterminate, there may be two equally adequate
computational interpretations of a process. What they have in common,
then, cannot be expressed from within the intentional stance: one needs
to leave the intentional stance and resort to some other stance. It turns
out that if Dennett wants to explain the difference between computing
mechanisms and other mechanisms, or explain what two adequate
interpretations of a computation have in common, he needs to individu-
ate computing mechanisms and their states using non-semantic lan-
Dennetts theory of content was very successful among philosophers
interested in CTM. For example, John Haugeland used a version of
Dennetts homuncular functionalism as an explication of the research
program, pursued by many in psychology and AI, of developing a CTM
(Haugeland 1978, 1985, 1997). For Haugeland, like for Dennett, the
content ascribed by a computational theory of a system, including a
CTM, came from the theorists interpretation of the system.
The author who elaborated Dennetts theory of content in the most
sophisticated and systematic way was perhaps Robert Cummins. Cum-
mins built a theory of computational explanation (1983), which included
a theory of computational content (1983, 1989), drawing from both
Dennett and Haugelands writings. Like Dennett and Haugelands theo-
394 Gualtiero Piccinini
ries, Cumminss theory was squarely based on the semantic view of
computation and explained content in terms of how an external observer
interpreted a system.
Unlike Dennett, who was frankly anti-realist about mental content,
Cummins took a more realist position (1983, 74-5). Cummins sharply
distinguished between mental content and computational content. He
argued that in formulating a CTM, psychologists and AI researchers
postulated (contentful) computational states as explanations for peoples
cognitive capacities, and offered his interpretational theory as an account
of the content of computational states postulated by CTM theorists. But
Cummins also argued that CTM fell short of explaining genuine mental
content, or as he put it, the intentionality of mental capacities. He
discussed five strategies to explain mental content in terms of computa-
tional content and argued that they all failed (1983, 91-100). Cummins
still stated that intentionality was somehow going to be accounted for in
terms of computation, but he added that he had no idea how this could
be done (1983, 89-90). He denied that he had a theory of mental content
(1989). So, Cumminss interpretational theory of content did nothing to
solve the philosophical problem of mental content with which we are
presently concerned (nor was it intended to do so). Moreover, Cum-
minss theory entailed that the same system could be interpreted in
different ways that ascribed different computations.
Another author in the interpretational semantics tradition is Patricia
Churchland. In her book written with neuroscientist Terrence Sejnowski,
she offered a detailed account of the brain as a computing mechanism,
predicated on the semantic view of computation. In explicating what it
meant to be a computer, Churchland and Sejnowski stated an informal
version of Cumminss theory (with a reference to a paper by Cummins
for more details; Churchland and Sejnowski 1992, 65). Unlike Cummins,
Churchland and Sejnowski did not discuss explicitly what notion of
content was at stake in their theory. Perhaps because of this, some
authors have accused Churchland and Sejnowski of being ambiguous as
to whether or not they ascribed genuine mental content to brains (Grush
2001). In light of the preceding discussion of interpretational semantics,
I offer a more charitable reading. Churchland and Sejnowski were
working with an interpretational semantics, which had the same advan-
tages and disadvantages of all interpretational semantics. On the one
hand, it applied equally well to organisms and artifacts. On the other
hand, it did not solve the problem of mental content. Churchland and
Sejnowski may choose whether to side with Dennett or Cummins: either
they accept Dennetts anti-realism about mental content, so that their
interpretational semantics explains mental content away (with Dennett),
or they endorse Cumminss realism about mental content, but then they
must defer to some other theory as far as mental content is concerned.
Functionalism, Computationalism, Mental Contents 395
Given that at least Churchland is in print denying the existence of
intrinsic intentionality (Churchland and Churchland 1983), presumably
she would side with Dennetts anti-realism about content.
In conclusion, interpretational semantics is a natural and attractive
way to cash out the semantic view of computation, but it comes at the
cost of losing the ability to explain how computing mechanisms differ
from other mechanisms, and what two distinct but equally adequate
computational interpretations of a process have in common. In order to
regain these abilities, an interpretational semanticist needs a non-seman-
tic way to individuate computational states. In addition, interpretational
semantics makes computational descriptions, even construed as seman-
tic descriptions, insufficient to characterize mental content as a real
property of minds. This may not trouble those who dont believe in
mental content to begin with, but it leads others to look elsewhere for a
theory of content.
VIII CTM meets Informational and
Teleological Semantics
Until the mid-1970s, Fodor freely appealed to the semantic view of
computation in formulating his version of CTM, without discussing the
need for a theory of content. In a series of papers in the late 1970s
(collected in Fodor 1981), he became more explicit about the relationship
between CTM and mental content. He argued that folk psychology
formulated its generalizations by quantifying over the semantic proper-
ties of propositional attitudes, whereas cognitive (computational) psy-
chology formulated its generalizations by quantifying over the syntactic
properties of mental representations. He added that the strength of a
computational psychology, postulating LOT, was that computational
psychology had the resources to reduce the generalizations of folk
psychology to scientific generalizations, so as to underwrite our intuitive
discourse about contentful propositional attitudes. This was because
computations, albeit being causally driven by the syntactic properties of
representations, were processes that (could) respect the semantic prop-
erties of representations. So, given a computational psychology, the two
stories about the mind, cognitive and folk, would eventually match, in
the sense that under the appropriate ascription of content to the states
posited by cognitive psychology, the relations between semantically
individuated folk psychological states would match the causal relation-
ships between cognitive psychological states. What was needed to com-
plete this picture was a theory ascribing the right contents to the
computational states: a theory of content (Fodor 1981; see also Fodor
1987, ch. 1).
396 Gualtiero Piccinini
A theory purporting to do this was FRS, but Fodor rejected it. He wrote
a critique of procedural semantics, a theory of content that was popular
in AI and psychology. According to procedural semantics, the content
of a computational instruction was given by the computational proce-
dures that executed that instruction. Since the relations between an
instruction and the computational procedures that operated on it were
functional relations, procedural semantics was a version of FRS. Fodor
argued that, since the procedures that executed computer instructions
were entirely internal to the machine and, when transformed into ma-
chine language, were naturally interpreted as referring to the shifting of
bits from one register of the machine to another, procedural semantics
reduced the content of computational descriptions to content about
shifting bits from one register to another, without ever involving any-
thing external to the machine (Fodor 1978). Besides procedural seman-
tics, Fodor also rejected FRS in general, in part because he saw that, given
a semantic view of computation, FRS was circular (e.g., see Fodor 1990,
ch. 1). Fodor maintained the semantic view of computation and treated
content ascription as prior to computation ascription. For him, compu-
tations were defined over representations, which were individuated by
their content, so computations were individuated by the semantic prop-
erties of the representations over which they were defined. Because of
this  which was the semantic view of computation  a theory of
content could not individuate contents by appealing to the notion of
computation, on pain of circularity.29
By the end of the 1970s, any computationalist philosopher of mind
who  like Fodor  took mental content as a real property of minds but
rejected FRS, needed an alternative (naturalistic) theory of content. This
demand was soon met by two new approaches, Informational Semantics
and Teleological Semantics (ITS). According to ITS, the content of a
Functionalism, Computationalism, Mental Contents 397
29 The most explicit discussion of this point that I know of is in Fodor 1998. He wrote
that since his notion of computation presupposed the notion of content, he could
not account for content in terms of computation (ibid., esp. 13). He also said
explicitly that because of this, his theory of content (unlike his theory of thought)
was not computational (ibid., 11). Notice that at least prima facie, the semantic view
of computation is consistent with the formality condition (Fodor 1980), according
to which computational processes are sensitive only to the formal (i.e., non-seman-
tic) properties of representations. The semantic view of computation is about how
to individuate computational states, whereas the formality condition is about which
properties of computational states are causally efficacious. So, according to Fodor,
[that] taxonomy in respect to content is compatible with the formality condition,
plus of minus a bit, is perhaps the basic idea of modern cognitive theory (Fodor
1980, 240, emphasis in original).
mental state came from natural relations between that state and the
minds environment. Informational Semantics said that the crucial rela-
tions for individuating content were informational, namely relations
determining what information was carried by a state (Dretske 1981,
1986). Teleological Semantics said that the crucial relations involved the
evolutionary history of the state, namely what the mechanism generat-
ing that state was selected for (Millikan 1984, 1993).30 Following Dretske
and Millikan, Fodor developed his own version of ITS (1987, 1990, 1998)
with the explicit goal of finding a theory of content for the repre-
sentations postulated by LOT.
The main difference between FRS and ITS is that while the former is
holistic and (partially) internalist, specifying (part of) the content of a
state by its relation to other contentful states, ITS is externalist and
atomistic, specifying the content of a state independently of its relation
to other contentful states. So, ITS specifies the content of a state inde-
pendently of what computations it enters.
According to Fodor, the combination of ITS and CTM offered our best
hope for a scientific theory of mind that would respect folk intuitions
about mental content, so that the stories told by cognitive and folk
psychology would match. This was because CTM accounted for how
mental processes could be causally efficacious while respecting the
semantic properties of mental representations, and ITS accounted for
how representations got their content (Fodor 1987, 1990, 1998).
Given a theory of content of the ITS form, it should be obvious that
being a computational state is insufficient for having mental content. I
can run an instantiation of the argument schema described in section V:
Premise 1. Having mental content is the same as entering certain
informational or teleological relations with the environment.
Premise 2. Being a computational state does not entail entering the
relations mentioned in premise 1.
Conclusion: Being a computational state does not entail having
mental content.
Premise 1 is just ITS. Premise 2 expresses the familiar fact that what we
ordinarily call computers, and use as computers, are rarely if ever
hooked up to the environment in the complicated ways postulated by
ITS. Some computing mechanisms, which computer scientists call em-
398 Gualtiero Piccinini
30 Incidentally, Millikan was another student of Sellars. At the beginning of her first
book (Millikan 1984), she stated that her theory of content was inspired by some of
Sellarss remarks.
bedded systems (e.g., cars computers and digital thermostats), are con-
nected to the environment in ways that resemble those postulated by
some versions of ITS, but they are not the typical case. Ordinarily,
whether something is a computing mechanism is independent of the ITS
relations it bears to the environment.
This conclusion would not come as a surprise to ITS theorists: ITS
theorists generally dont ascribe mental content to ordinary computers.31
This has the important consequence that, to the extent that they accept
ordinary computation ascription, ITS theorists who believe in CTM are
committed to there being something in common between computing
mechanisms that satisfy the demands of ITS for mental content and
ordinary (non-embedded) computing mechanisms: although they are all
computing mechanisms, some have mental content while others dont.
That is, a consequence of conjoining CTM and ITS is that minds and
ordinary computing mechanisms have something in common that can-
not be specified by ITS. The way to specify it, I submit, is by a non-se-
mantic account of computational states.
To summarize, given ITS the search for a theory of mental content is
not by itself a motivation to endorse CTM, because the solution to the
problem of mental content is independent of CTM. If anything, it is the
search for a theory of contentful mental states that still motivates com-
putationalist philosophers who want to save the semantic view of com-
putation to match CTM with ITS. If they succeed, they find themselves
in a position from which they cannot tell what minds have in common
with ordinary computing mechanisms. In order to tell, they need a
non-semantic way to individuate computing mechanisms and their
IX CTM meets Intentional Eliminativism
All the computationalist philosophers discussed until now shared the
semantic view of computation. The first who questioned their assump-
tion was Stephen Stich (1983). Like other computationalist philosophers,
Stichs primary goal was not to give a philosophical account of compu-
tation but a theory of mind; his rejection of the semantic view of compu-
tation was an implicit consequence of his theory of mind. Stich was
motivated by the belief that folk psychology, including the mental
contents it postulated, would be eliminated in favor of a cognitive
Functionalism, Computationalism, Mental Contents 399
31 For a critical discussion of this feature of Dretske and Fodors view, see Dennett
1987, ch. 8.
psychological theory of mind. Contra Fodor, he argued that the generali-
zations of cognitive psychology would not match those of folk psychol-
ogy (Stich 1983, chs. 4, 7, and 9).
Stich formulated a version of CTM that did not require mental states
to have content, a theory that he called syntactic theory of mind. In order
to have a CTM without mental content, Stich implicitly rejected the
semantic view of computation. According to him, the mind was compu-
tational, but computational descriptions did not ascribe semantic prop-
erties. For something to be computational, its physical states must be
mappable onto syntactically defined states, without presupposing any
semantics. A system was a computing mechanism if and only if there
was a mapping between its behaviorally relevant physical states and a
class of syntactic types, specified by a grammar that defined how com-
plex types could be formed out of primitive types. According to Stich,
the mind was a computing mechanism in this sense (Stich 1983).
By formulating his version of CTM in terms of his syntactic view of
computation and doing away with mental content, Stich renounced
what Fodor considered the crucial consideration in favor of CTM: the
hope that cognitive psychological generalizations ranging over syntac-
tically individuated states would correspond to folk psychological gen-
eralizations ranging over contentful propositional attitudes. According
to Fodor, the point of CTM was to explain how a mere mechanism could
mirror semantic relations by invoking the match between mechanical
processes that responded only to syntactic properties (computations)
and processes individuated by semantic properties (inferences). From
Fodors point of view, if one followed Stich in denying that mental states
had content, it was unclear why and how mental states and their func-
tional relations should be construed as being computational. Perhaps
because of this, Stichs syntactic theory of mind won few converts. But
Stich argued that a syntactic theory of mind offered the best explanation
of mental phenomena: for instance, Stich said that beliefs in the folk sense
would likely be identified with some non-contentful, syntactically indi-
viduated states that had a functional role similar to the one played by
beliefs in folk psychology (Stich 1983, ch. 11). Moreover, Stich argued
that his syntactic theory of mind was the best construal of the computa-
tional theories of mental phenomena offered by cognitive psychologists.
If Stichs proposal of a syntactic theory of mind is coherent, it shows
how CTM can be formulated without the semantic view of computation.
This point is independent of Stichs intentional eliminativism. There is
an important sense in which Stichs syntactic theory of mind is compat-
ible with the existence of mental content and the Fodorian argument for
CTM. Stichs syntactic criterion for mental states can be taken as a way
to individuate mental states and processes as computational in a non-se-
mantic way, while leaving open the question of whether they also have
400 Gualtiero Piccinini
content and whether there are generalizations ranging over contentful
states that match those formulated over syntactically individuated
This is an open possibility so long as we abandon the semantic view
of computation, which maintains that computational states are individu-
ated by their content. For if computational states are individuated by
their content, it would be impossible to individuate them non-semanti-
cally, as Stichs theory requires, and then ask whether they have content
and what content they have. From the point of view of the semantic view
of computation, Stichs coupling of CTM and intentional eliminativism,
according to which mental states are computational but contentless, is
incoherent. And from the point of view of Stichs combination of CTM
and intentional eliminativism, the semantic view of computation, and
any version of CTM that is formulated using the semantic view of
computation, begs the question of whether the computational mind has
Indeed, Stichs notion of syntax has been challenged on the grounds
that it makes no sense to speak of syntax without semantics. According
to this line of thought, something can be a token of a syntactic type only
relative to a language in which that token has content (e.g., Crane 1990;
Jacquette 1991). If this is right, then Stichs proposal is incoherent, but I
dont think it is.
The coherence of Stichs proposal is easy to see when we reflect on the
functional properties of stored-program computers.33 Some special
mechanisms, namely stored-program computers, have the ability to
respond to (non-semantically individuated) strings of tokens stored in
their memory by executing sequences of primitive operations, which in
turn generate new strings of tokens that get stored in memory. Different
bits and pieces of these strings of tokens have different effects on the
machine. Because of this, the strings of tokens can be analyzed into
sub-strings. An accurate description of how tokens can be compounded
into sub-strings, and sub-strings can be compounded into strings, which
does not presuppose that the strings of tokens have any content, may be
called the syntax of the system of strings manipulated by the computer.
Some strings, called instructions, have the function of determining, at
any given time, which operations are to be performed by the computer
Functionalism, Computationalism, Mental Contents 401
32 In fact, Frances Egan has advocated the conjunction of a non-semantically formu-
lated CTM, a la Stich, with the view that computational states have content (Egan
1999, 181).
33 For a more detailed account of stored-program computers along these lines, see
Piccinini 2003a, ch. 10.
on the input strings. Because of how computers are designed, the global
effect of an instruction on the machine can be reduced to the effects of
its sub-strings on the machine. Then, the effect of sub-strings on the
computer can be assigned to them as their content, and the way in which
the content of the whole string depends on the content of its sub-strings
can be specified by recursive clauses, with the result that the global effect
of a string on the computer is assigned to it as its content. This assignment
constitutes an internal semantics of a computer. An internal semantics
assigns as contents to a system its own internal components and activi-
ties, whereas an ordinary (external) semantics assigns as contents to a
system objects and properties in the systems environment.34 Given that
the strings manipulated by a computer may have a syntax (which
determines how they are manipulated), and some of them have an
internal semantics, they may be called a language, and indeed that is
what computer scientists call them. None of this entails that computer
languages have any external semantics, i.e. any content in the sense used
by Stichs critics, although it is compatible with their having one. Stich
may be construed as arguing that the functional organization of the mind
is similar to that of a stored-program computer, so that the mind contains
a system of strings of tokens with a syntax analogous to that of the strings
manipulated by stored-program computers. Stich would probably have
no difficulty in accepting that if the brain is capable of storing and
executing its own instructions, then some of the mental strings also have
an internal semantics.
The above account of syntax is functional, specified in terms of the
components of a stored-program computer, their states, and their inter-
actions. From the vantage point of this functional view of computation,
not only do we see the coherence of Stichs proposal, but we can also give
a functional account of his notion of syntax without presupposing any
external semantics.
Stichs proposal shows that one can be a computationalist without
having a theory of content and while rejecting the semantic view of
computation, because one can be a computationalist without believing
in mental content at all. A computationalist who wishes not to beg the
question against the intentional eliminativist should formulate CTM
without the semantic view of computation, and independently of any
theory of content.
402 Gualtiero Piccinini
34 For more on internal vs. external semantics, see Fodor 1978; Dennett 1987; and
Piccinini 2003a, chs. 9 and 10.
X CTM With or Without Semantics
The first moral of this paper is that CTM was originally conceived in
tandem with the semantic view of computation, and this convinced a lot
of philosophers that CTM necessarily ascribed content to the mind.
Between the 1940s and the late 1960s, both in science and philosophy,
computationalists promoted CTM not only as offering a mechanistic
explanation of the mind but also, by construing computational descrip-
tion semantically, as offering the beginning of a naturalistic account of
In the 1970s, it became clear that CTM per se offered no solution to the
problem of mental content. The ensuing investigations of a theory of
content revealed four main ways to combine CTM with a theory of
content. The first combines CTM with Functional Role Semantics (FRS),
which sees mental content as (partially) reducible to the computational
relations among mental states. This option cannot construe computa-
tional relations semantically on pain of circularity, and hence it presup-
poses a non-semantic way of individuating computational states. The
second combines CTM with interpretational semantics, which sees men-
tal content as a tool for predicting behavior. This option maintains the
semantic view of computation at the cost of denying the reality of mental
content. The third combines CTM with Informational or Teleological
Semantics (ITS), which sees content as reducible to a combination of
causal and counterfactual relations between mental states and the envi-
ronment. This option maintains the semantic view of computation at the
cost of being inapplicable to ordinary computing mechanisms, because
most ordinary computing mechanisms dont enter the causal and coun-
terfactual relations postulated by ITS. The fourth combines CTM with
intentional eliminativism, which abandons the semantic view of compu-
tation in favor of a non-semantic construal of computation.
These options seem to exhaust the possibilities that are open to the
computationalist: either mental content comes from the computations
themselves (FRS), or it comes from some non-computational natural
properties of the content-bearing states (ITS), or it is in the eye of the
beholder (interpretational semantics), or there is no mental content at all
(eliminativism). Under any of these options, either the problem of mental
content is solved by something other than computation ascription, or
computation ascription must be construed non-semantically, or mental
content is unreal. Usually more than one of the above is true of each
For each of the above theories of content, neither the theory of content
entails CTM nor CTM entails the theory of content. None of the existing
theories of mental content offers a reason to endorse CTM. Whether the
mind is computational and whether the mind has content (and how it
Functionalism, Computationalism, Mental Contents 403
manages to have content) are different problems that need to be solved
independently of each other. The semantic view of computation also
begs the question of the FRS and intentional eliminativist theorists, who
need a non-semantic individuation of computational states in order to
formulate their views. In order to keep the two problems separate, we
should avoid formulating CTM as a theory that ascribes content to the
mind, as it is often done (e.g., by Fodor 1998 and Horst 2003). Even those
who are skeptical about full-blown mental content but believe in some
form of computational content (e.g., Churchland and Sejnowski 1992)
should avoid formulating CTM as a theory that ascribes content to the
mind. CTM should be formulated in a way that is neutral about content,
leaving it to considerations about content to determine which theory of
content is correct. CTM is a theory of the internal mechanisms of the
mind or brain, which may or may not explain some mental phenomena.
So, the semantic view of computation should be abandoned in favor of
the functional view of computation, and CTM should be formulated
without using semantic language. Fortunately, as I argue on inde-
pendent grounds in Piccinini 2003a, chs. 9 and 10, this is also the best
way to understand computing mechanisms in their own right. Stich is
right in one important respect: in order to understand computing mecha-
nisms and how they work (as opposed to why they are built and how
they are used), there is no need to invoke content; its actually misleading
to do so.
Construing CTM without the semantic view of computation leaves the
field entirely open for different positions about content. Perhaps some
computational states have content and some dont; perhaps all do or
none do. Perhaps some have content in one sense and not in others. CTM
should not be seen as answering any of these questions. If the mind is
computational but has no content, then CTM will explain the mind
without requiring a theory of content. If the mind does have content,
then this is going to be explained by a theory of content. If mental states
are both contentful and computational, then the true version of CTM and
the true theory of content will be compatible with each other. One
example of compatibility is offered by combining a non-semantically
formulated version of LOT with ITS; this is analogous to Fodors view
minus his semantic view of computation. Another example is the com-
bination of a non-semantically formulated version of CTM and interpre-
tational semantics; this is analogous to Dennetts view minus his
semantic view of computation. A third example is the conjunction of a
non-semantically formulated version of CTM with FRS.
404 Gualtiero Piccinini
XI Two Consequences
If questions of content are independent of questions of computation,
there are some consequences that deserve to be explored. I will briefly
mention two:
1. During the last two decades, it has become common to hear criti-
cisms of CTM based on the rejection of representationalism (Brooks 1997;
Thelen and Smith 1994; van Gelder 1995; and certain passages in Clark
1997). According to these criticisms, some or all mental phenomena can
be explained without postulating contentful mental states, and therefore
CTM should be rejected. As Ive tried to show, many computationalists
have endorsed the semantic view of computation, and therefore their
position is vulnerable to this criticism. But I also argued that the semantic
view of computation should be rejected. If this is done, then the anti-rep-
resentationalist critique of CTM turns out to be confused in the same way
that the semantic view of computation is. Even if we dont need repre-
sentations to explain cognition (which I doubt), this would do nothing
to undermine CTM per se, but only the combination of CTM with
representationalism. CTM can and should be formulated independently
of any theory of content, which makes it invulnerable to anti-repre-
sentationalist critiques.
2. Above, I mentioned Fodors argument according to which CTM is
our best theory of mind because, by postulating causal processes that
can mirror semantic relations between representations, it offers the hope
to generate a scientific theory of mind close to our folk theory of mind.
(Fodor has forcefully made this argument in conjunction with ITS, but a
version of it could be run in conjunction with FRS.) However, if we accept
that the question of computation is independent of the question of
content, it becomes clear that this argument is missing a crucial premise.
Before we accept that CTM has the potential to match contentful rela-
tions with computational processes, we should ask by what mechanism
this match is achieved. In other words, we need to conjoin CTM not only
with a theory of content, but also with a theory of how the computational
relations get to match the semantic properties of the internal states.
Notice that the mechanism that accomplishes the matching cannot be
computational on pain of circularity. For if it were a computing mecha-
nism, we should ask whether its computational processes match its
semantic properties. If they dont, then it is unclear how such a mecha-
nism could achieve the syntax-semantics match in the first mechanism.
If they do, we need to answer the question of how they do, and we are
back where we started. So the matching must be done by a non-compu-
tational mechanism. What mechanism is it and how does it work? At
some point, Fodor formulated a problem very similar to this, which he
called the coordination problem, and argued that its solvable (Fodor
Functionalism, Computationalism, Mental Contents 405
1994, 12ff., 86). More recently, he has come close to admitting that he
doesnt know how to solve this problem (see Fodor 2000, esp. 71-8).
Without a solution, his argument for CTM doesnt go through. This may
be one of the reasons for his skepticism that CTM is going to offer a
complete explanation for the (non-conscious aspect of the) mind (Fodor
Received: October 2003
Revised: March 2004
Block, N. 1986. Advertisement for a Semantics for Psychology, in Midwest Studies in
Philosophy X: Studies in the Philosophy of Mind, P. French, T.E. Uehling, jr. and H.K.
Wettstein, eds. Minneapolis: University of Minnesota Press.
Brandom, R.B. 1994. Making it Explicit: Reasoning, Representing, and Discursive Commitment.
Cambridge, MA: Harvard University Press.
Brooks, R.A. 1997. Intelligence without Representation, in Mind Design II, J. Haugeland,
ed. Cambridge, MA: MIT Press.
Chomsky, N. 1957. Syntactic Structures. The Hague: Mouton.
______. 1965. Aspects of a Theory of Syntax. Cambridge, MA: MIT Press.
Churchland, P.M. 1979. Scientific Realism and the Plasticity of Mind. Cambridge: Cambridge
University Press.
______. 1989. A Neurocomputational Perspective. Cambridge, MA: MIT Press.
Churchland, P.M. and P.S. Churchland. 1983. Stalking the Wild Epistemic Engine. Noûs
17 5-18.
Churchland, P.S. and T.J. Sejnowski. 1992. The Computational Brain. Cambridge, MA: MIT
Clark, A. 1997. Being There. Cambridge, MA: MIT Press.
Crane, T. 1990. The Language of Thought: No Syntax Without Semantics. Mind and
Language 5 187-212.
Cummins, R. 1983. The Nature of Psychological Explanation. Cambridge, MA: MIT Press.
______. 1989. Meaning and Mental Representation. Cambridge, MA: MIT Press.
Dennett, D.C. 1969. Content and Consciousness. London: Routledge and Kegan Paul.
______. 1971. Intentional Systems. Journal of Philosophy 68 87-106.
______. 1978. Brainstorms. Cambridge, MA: MIT Press.
______. 1987. The Intentional Stance. Cambridge, MA: MIT Press.
Deutsch, J.A. 1960. The Structural Basis of Behavior. Chicago: University of Chicago Press.
Dretske, F.I. 1981. Knowledge and the Flow of Information. Cambridge, MA: MIT Press.
406 Gualtiero Piccinini
______. 1986. Misrepresentation, in Belief: Form, Content, and Function, R. Bogdan, ed. New
York: Oxford University Press.
Egan, F. 1999. In Defence of Narrow Mindedness. Mind and Language 14 177-194.
Field, H. 1978. Mental Representation. Erkenntnis 13 9-61.
Fodor, J.A. 1965. Explanations in Psychology, in Philosophy in America, M. Black, ed.
London: Routledge and Kegan Paul.
______. 1968a. Psychological Explanation. New York: Random House.
______. 1968b. The Appeal to Tacit Knowledge in Psychological Explanation. Journal of
Philosophy 65 627-40.
______. 1975. The Language of Thought. Cambridge, MA: Harvard University Press.
______. 1978. Tom Swift and His Procedural Grandmother. Cognition 6. Reprinted in
Fodor 1981.
______. 1980. Methodological Solipsism Considered as a Research Strategy in Cognitive
Psychology. Behavioral and Brain Sciences 3. Reprinted in Fodor 1981.
______. 1981. Representations. Cambridge, MA: MIT Press.
______. 1987. Psychosemantics. Cambridge, MA: MIT Press.
______. 1990. A Theory of Content and Other Essays. Cambridge, MA: MIT Press.
______. 1994. The Elm and the Expert: Mentalese and Its Semantics. Ca mbr idge , MA: MIT Pr ess.
______. 1998. Concepts. Oxford: Clarendon Press.
______. 2000. The Mind Doesnt Work That Way. MIT Press: Cambridge, MA.
Geach, P.T. 1956. Mental Acts. London: Routledge & Paul.
Grush, R. 2001. The Semantic Challenge to Computational Neuroscience, in Theory and
Method in the Neurosciences, P. Machamer, R. Grush and P. McLaughlin, eds. Pitts-
burgh, PA: University of Pittsburgh Press.
Harman, G. 1968. Three Levels of Meaning. Journal of Philosophy 65 590-602.
______. 1970. Sellars Semantics. The Philosophical Review 79 404-419.
______. 1973. Thought. Princeton: Princeton University Press.
______. 1988. Wide Functionalism, in Cognition and Representation, S. Schiffer and S. Steele,
eds. Boulder: Westview.
Haugeland, J. 1978. The Nature and Plausibility of Cognitivism. Behavioral and Brain
Sciences 2 215-60.
______. 1985. Artificial Intelligence: The Very Idea. Cambridge, MA: MIT Press.
______. 1997. Mind Design II. Cambridge, MA: MIT Press.
Horst, S. 2003. The Computational Theory of Mind, in The Stanford Encyclopedia of
Philosophy (Fall 2003 Edition), ed. E. N. Zalta. URL = <
Ince, D., ed. 1992. Mechanical Intelligence. The Collected Works of Alan Turing. Amsterdam:
Functionalism, Computationalism, Mental Contents 407
Jackson, F. and P. Pettit 1988. Functionalism and Broad Content. Mind 97 381-400.
Jacquette, D. 1991. The Myth of Pure Syntax, in Topics in Philosophy and Artificial Intelli-
gence, L. Albertazzi and R. Poli, eds. Bozen: Istituto Mitteleuropeo di Cultura.
Loar, B. 1981. Mind and Meaning. Cambridge: Cambridge University Press.
Loewer, B. and G. Rey, eds. 1991. Meaning in Mind: Fodor and his Critics. Oxford: Blackwell.
Marr, D. 1982. Vision. New York: Freeman.
McCorduck, P. 1979. Machines Who Think: A Personal Inquiry into the History and Prospects
of Artificial Intelligence. San Francisco, CA: Freeman.
McCulloch, W.S. 1965. Embodiments of Mind. Cambridge, MA: MIT Press.
McCulloch, W.S. and W.H. Pitts. 1943. A Logical Calculus of the Ideas Immanent in
Nervous Activity. Bulletin of Mathematical Biophysics 7 115-133.
Miller, G.A., E.H. Galanter, and K.H. Pribram. 1960. Plans and the Structure of Behavior. New
York: Holt.
Millikan, R.G. 1984. Language, Thought, and Other Biological Categories: New Foundations for
Realism. Cambridge, MA: MIT Press.
______. 1993. White Queen Psychology and Other Essays for Alice. Cambridge, MA: MIT Press.
Oppenheim, P. and H. Putnam 1958. Unity of Science as a Working Hypothesis, in
Minnesota Studies in the Philosophy of Science, Volume II Concepts, Theories, and the
Mind-Body Problem, H. Feigl, M. Scriv en and G. Maxwell, eds. Minneapolis: Univer-
sity of Minnesota Press.
Peacocke, C. 1992. A Study of Concepts. Cambridge, MA: MIT Press.
Piccinini, G. 2000. Turings Rules for the Imitation Game. Minds and Machines 10 573-82.
______. 2002. Review of John von Neumanns The Computer and the Brain. Minds and
Machines 12 449-53.
______. 2003a. Computations and Computers in the Sciences of Mind and Brain. Doctoral
Dissertation, Pittsburgh, PA: University of Pittsburgh. URL = <>
______. 2003b. Alan Turing and the Mathematical Objection. Minds and Machines 12 23- 48.
______. forthcoming a. Functionalism, Computationalism, and Mental States. Studies in
the History and Philosophy of Science.
______. forthcoming b. The First Computational Theory of Mind and Brain: A Close Look
at McCulloch and Pittss Calculus of Ideas Immanent in Nervous Activity.
Putnam, H. 1960. Minds and Machines, in Dimensions of Mind: A Symposium, S. Hook,
ed. New York: Collier.
______. 1963. Brains and Behavior, in Analytical Philosophy, R.J. Butler, ed. New York:
Barnes and Noble. Reprinted in N. Block, ed., Readings in Philosophy of Psychology,
Volume 1: 24-36. London: Methuen 1980.
______. 1964. Robots: Machines or Artificially Created Life? Journal of Philosophy 61 66 8-9 1.
Reprinted in H. Putnam, Mind, Language and Reality: Philosophical Papers, Volume 2.
Cambridge: Cambridge University Press 1975.
408 Gualtiero Piccinini
______. 1967a. The Mental Life of Some Machines, in Intentionality, Minds, and Perception,
H.-N. Castañeda, ed. Detroit: Wayne State University Press.
______. 1967b. Psychological Predicates, in Art, Philosophy, and Religion. Pittsburgh, PA:
University of Pittsburgh Press. Reprinted as The Nature of Mental States in W.
Lycan, ed., Mind and Cognition: An Anthology. Second Edition, Malden: Blackwell
______. 1974. Comments on Wilfrid Sellars. Synthese 27 445-55.
______. 1988. Representation and Reality. Cambridge, MA: MIT Press.
______. 1997. A Half Century of Philosophy, Viewed from Within. Daedalus (Winter 1997)
Pylyshyn, Z.W. 1984. Computation and Cognition. Cambridge, MA: MIT Press.
Ryle, G. 1949. The Concept of Mind. London: Hutchinson.
Sellars, W. 1954. Some Reflections on Language Games. Philosophy of Science 21 204-28.
Reprinted in Sellars 1963.
______. 1956. Empiricism and the Philosophy of Mind, in Minnesota Studies in the Philoso-
phy of Science, Vol. I, The Foundations of Science and the Concepts of Psychology and
Psychoanalysis, H. Feigl and M. Scriven, eds. Minneapolis: University of Minnesota
Press. Reprinted in Sellars 1963.
______. 1961. The Language of Theories, in Current Issues in the Philosophy of Science, H.
Feigl and G. Maxwell, eds. New York: Holt, Rinehart, and Winston. Reprinted in
Sellars 1963.
______. 1963. Science, Perception, and Reality. Atascadero: Ridgeview.
______. 1967 Science and Metaphysics: Variations on Kantian Themes. London: Routledge and
Kegan Paul.
______. 1974. Meaning as Functional Classification. Synthese 27 417-37.
______. 1975. Autobiographical Reflections, in Action, Knowledge, and Reality: Studies in
Honor of Wilfrid Sellars, H.-N. Castañeda, ed. Indianapolis: Bobbs-Merrill.
Shannon, C.E. and J. McCarthy. 1956. Automata Studies. Princeton, NJ: Princeton University
Smith, B.C. 1996. On the Origin of Objects. Cambridge, MA: MIT Press.
Stich, S. 1983. From Folk Psychology to Cognitive Science. Cambridge, MA: MIT Press.
Sutherland, N.S. 1960. Theories of Shape Discrimination in Octopus. Nature 186 840-4.
Thelen, E. and L. Smith 1994. A Dynamic Systems Approach to the Development of Cognition
and Action. Cambridge, MA: MIT Press.
Turing, A.M. (1936-7 [1965]). On Computable Numbers, with an Application to the
Entscheidungsproblem. In The Undecidable, M. Davis, ed. Ewlett: Raven.
______. 1950. Computing Machinery and Intelligence. Mind 59 433-60.
van Gelder, T. 1995. What Might Cognition Be, if not Computation? The Journal of
Philosophy 92 345-81.
Vendler, Z. 1972. Res Cogitans: An Essay in Rational Psychology. Ithaca: Cornell University
Functionalism, Computationalism, Mental Contents 409
von Neumann, J. 1958. The Computer and the Brain. New Haven: Yale University Press.
Wiener, N. 1948. Cybernetics or Control and Communication in the Animal and the Machine.
Cambridge, MA: MIT Press.
Wittgenstein, L. 1953. Philosophical Investigations. New York: Macmillan.
410 Gualtiero Piccinini
... Turing (1936) work was a significant step in that he managed to conceptualise an abstract, automatic machine that took care of information processing tasks, in particular mathematical operations. However, the work was otherwise unspecified in terms of the contents of information processing, because the symbols in the machine and related operation could be defined in any way, respecting certain constraints (Piccinini, 2004). The Turing machine was both limited and enabled by the level of abstraction at which Turing operated, namely computational processes (Saariluoma & Rauterberg, 2016). ...
... Computation is thus approached pragmatically, as the ground of computers and thus the target "as we find it". There is no pressing need to take a stand on the ontological status of computation with respect to mental processes, i.e., whether they are constituted as such (Clark, 1996;Dreyfus, 2007;Lucas, 1996;Newell & Simon, 1976;Penrose, 1990;Piccinini, 2004;Rapaport, 2012;Simon, 1995), other than to note that if the target is understood as computational information processing, then over the mapping relationship, any representations of the source need to land eventually in the computational idiom. This makes, for example, computational cognitive science a prima facie suitable theory language for cognitive mimetics, where the empirical question is not whether cognition is computation, but whether it is computable (Rapaport, 2012). ...
... A problem in cognitive science and AI has been that the contents of the symbols do not enter into the computational processes as such (Harnad, 1990;Sayre, 1987). In the philosophy of mind and cognition, the problem of computation and contents has arrived at a similar conclusion, at least according to Piccinini (2004), that questions of computation are independent of questions of content. In the simplistic addition example, the computational process runs the same way even if we assign it to count apples, oranges or money, and this happens not to be a problem since the operation of addition is abstract and applies to all cases of numerical addition. ...
... The model then went on to revolutionize computational theory, inspiring von Neumann and thus the digital computing legacy we have inherited (Cobb, 2020). Neuroscientists at the time of McCullough and Pitts' writing however noted that the model of nervous activity they schematized was in fact substantially different from how the nervous system is actually organized (Cobb, 2020;Piccinini, 2004). Subsequent attempts to identify logic gates in nervous tissue have had some success, while also demonstrating that their operation is different from McCullough and Pitts' formulations (Dobosiewicz et al., 2019). ...
Full-text available
The brain-as-computer metaphor has anchored the professed computational nature of mind, wresting it down from the intangible logic of Platonic philosophy to a material basis for empirical science. However, as with many long-lasting metaphors in science, the computer metaphor has been explored and stretched long enough to reveal its boundaries. These boundaries highlight widening gaps in our understanding of the brain’s role in an organism’s goal-directed, intelligent behaviors and thoughts. In search of a more appropriate metaphor that reflects the potentially noncomputable functions of mind and brain, eight author groups answer the following questions: (1) What do we understand by the computer metaphor of the brain and cognition? (2) What are some of the limitations of this computer metaphor? (3) What metaphor should replace the computational metaphor? (4) What findings support alternative metaphors? Despite agreeing about feeling the strain of the strictures of computer metaphors, the authors suggest an exciting diversity of possible metaphoric options for future research into the mind and brain.
... The model then went on to revolutionize computational theory, inspiring von Neumann and thus the digital computing legacy we have inherited (Cobb, 2020). Neuroscientists at the time of McCullough and Pitts' writing however noted that the model of nervous activity they schematized was in fact substantially different from how the nervous system is actually organized (Piccinini, 2004). Subsequent attempts to identify logic gates in nervous tissue have had some success, while also demonstrating that their operation is different from McCullough and Pitts' formulations (Dobosiewicz et al., 2019). ...
Full-text available
The computer metaphor derives from two parallel histories of thought, the first questioning the operation and activity of minds and brains and the second aiming at finding a minimal model of the mind and brain. These histories informed separate hypotheses of mind, epistemic constructivism, and the mechanistic hypothesis; we focus on the latter. First, we briefly review how the brain has historically been discussed with machine metaphors and identify five tenets that define a machine. We review findings in neuroscience that motivate that the brain demonstrates exceptions to these tenets and thus ought not to be considered a machine. We offer that an alternative classification may be found in far-from-equilibrium self-organizing systems known as dissipative structures. We review the properties of these systems and suggest that the brain is more like a dissipative structure than a machine. If brains do not fit the mechanistic hypothesis underwriting the computer metaphor, then the cognitive sciences may need to seek alternative metaphors based on the assumption that minds and brains are other natural systems, namely dissipative structures.
... Fodor (1983) is one of the forerunners of the representationalist approach based on the modularity of the mind, as he asserts the impossibility of genuine conceptual learning outside the set of pre-existing representations. Representationalism is based on the idea that cognition operates on the basis of representations of intrinsic functionality called internalism or objective representations of a world independent of the subject called externalism (Piccinini, 2004). There is also the idea of computationalism in which cognition is reduced to processes and operations performed on entities that carry some form of content, functionally structured from input and output channels of information (Sprevak, 2010). ...
Conference Paper
Full-text available
Our paper aims to provide a proof of concept about the theoretical framework of enactive and ecological approaches to the field of perceptual learning of mathematics with digital technology. We reporst on the finger gnosis or finger knowledge that school children deploy when engaging with digital technologies such as Touchcounts and Rakin. From our theoretical lens we contrast both applications and conclude that they offer rich possibilities for number learning, however, Touchcounts adheres better with enactive and ecological foundations by prompting finger gnosis in different ways, while the Rakin technology application restricts actions and gestures, leaning toward representationalist cognitivism.
... Thus, simply adding Cummins's interpretational semantics to the deflationary account might make it viciously circular (cfr. Piccinini 2004). 6 Just to give and example, if F* takes as arguments Jackson (7 th president) and Harrison (9 th president), it will yield Lincoln as value (16 th president). ...
Full-text available
The deflationary account of representations purports to capture the explanatory role representations play in computational cognitive science. To this end, the account distinguishes between mathematical contents, representing the values and arguments of the functions cognitive devices compute, and cognitive contents, which represent the distal states of affairs cognitive systems relate to. Armed with this distinction, the deflationary account contends that computational cognitive science is committed only to mathematical contents, which are sufficient to provide satisfactory cognitive explanations. Here, I scrutinize the deflationary account, arguing that, as things stand, it faces two important challenges deeply connected with mathematical contents. The first depends on the fact that the deflationary account accepts that a satisfactory account of representations must deliver naturalized contents. Yet, mathematical contents have not been naturalized, and I claim that it is very doubtful that they ever will. The second challenge concerns the explanatory power of mathematical contents. The deflationary account holds that they are always sufficient to provide satisfactory explanations of cognitive phenomena. I will contend that this is not the case, as mathematical contents alone are not sufficient to explain why deep neural networks misclassify adversarial examples.
... Functionalist usually construe this problem in terms of a difference between a semantics that it "internal" (Piccinini, 2004(Piccinini, , 2006; the "syntactic" semantics of Rappaport) and that which is "external" (semantics proper). As Piccinini (2004, p. 402) puts it 14 : ...
Full-text available
The heyday of discussions initiated by Searle's claim that computers have syntax, but no semantics has now past, yet philosophers and scientists still tend to frame their views on artificial intelligence in terms of syntax and semantics. In this paper I do not intend to take part in these discussions; my aim is more fundamental, viz. to ask what claims about syntax and semantics in this context can mean in the first place. And I argue that their sense is so unclear that that their ability to act as markers within any disputes on artificial intelligence is severely compromised; and hence that their employment brings us nothing more than an illusion of explanation.
Current orthodoxy takes representation to be essential to computation. However, a philosophical account of computation that does not appeal to representation would be useful, given the difficulties involved in successfully theorizing representation. Piccinini’s recent mechanistic account of computation proposes to do just that: it couches computation in terms of what certain mechanisms do without requiring the manipulation or processing of representations whatsoever (Piccinini, 2015). Most crucially, mechanisms must process medium-independent vehicles. There are two ways to understand what "medium-independence" means on this account; however, on either understanding, the account fails. Either too many things end up being counted as computational, or purportedly natural computations (e.g., neural computations) cannot be counted at all. In the end, illustrating this failure sheds some light on the way to revise the orthodoxy in the hope of a better account of computation.
I argue that good metaphysics and good metascience go hand in hand and go on to clarify my egalitarian ontology of levels, defend the aspect view of realization, insist that computation must be medium independent and neurocognitive processes are largely computational even though some medium-dependent phenomena affect neural computation, and sketch an account of mental representation in terms of neural representation.
Full-text available
Situated approaches to cognition maintain that cognition is embodied, embedded, enactive, and affective (and extended, but that is not relevant here). Situated approaches are often pitched as alternatives to computational and representational approaches, according to which cognition is computation over representations. I argue that, far from being opposites, situatedness and neural representation are more deeply intertwined than anyone suspected. To show this, I introduce a neurocomputational account of cognition that relies on neural representations. I argue not only that this account is compatible with (non-question-begging) situated approaches, but also that it requires embodiment, embeddedness, enaction, and affect at its very core. That is, constructing neural representations and their semantic content, and learning computational processes appropriate for their content, requires a tight dynamic interaction between nervous system, body, and environment. Most importantly, I argue that situatedness is needed to give a satisfactory account of neural representation: neurocognitive systems that are embodied, embedded, affective, dynamically interact with their environment, and use feedback from their interaction to shape their own representations and computations (1) can construct neural representations with original semantic content, (2) their neural vehicles and the way they are processed are automatically coordinated with their content, (3) such content is causally efficacious, (4) is determinate enough for the system's purposes, (5) represents the distal stimulus, and (6) can misrepresent. This proposal hints at what is needed to build artifacts with some of the basic cognitive capacities possessed by neurocognitive systems.
Full-text available
Computationalism says that brains are computing mechanisms, that is, mechanisms that perform computations. At present, there is no consensus on how to formulate computationalism precisely or adjudicate the dispute between computationalism and its foes, or between different versions of computationalism. An important reason for the current impasse is the lack of a satisfactory philosophical account of computing mechanisms. The main goal of this dissertation is to offer such an account. I also believe that the history of computationalism sheds light on the current debate. By tracing different versions of computationalism to their common historical origin, we can see how the current divisions originated and understand their motivation. Reconstructing debates over computationalism in the context of their own intellectual history can contribute to philosophical progress on the relation between brains and computing mechanisms and help determine how brains and computing mechanisms are alike, and how they differ. Accordingly, my dissertation is divided into a historical part, which traces the early history of computationalism up to 1946, and a philosophical part, which offers an account of computing mechanisms. The two main ideas developed in this dissertation are that (1) computational states are to be identified functionally not semantically, and (2) computing mechanisms are to be studied by functional analysis. The resulting account of computing mechanism, which I call the functional account of computing mechanisms, can be used to identify computing mechanisms and the functions they compute. I use the functional account of computing mechanisms to taxonomize computing mechanisms based on their different computing power, and I use this taxonomy of computing mechanisms to taxonomize different versions of computationalism based on the functional properties that they ascribe to brains. By doing so, I begin to tease out empirically testable statements about the functional organization of the brain that different versions of computationalism are committed to. I submit that when computationalism is reformulated in the more explicit and precise way I propose, the disputes about computationalism can be adjudicated on the grounds of empirical evidence from neuroscience.